
Nonlinear Estimation of External Power System
Dynamic Equivalent Parameters
R. Gueddouche1
Abstract-- Based on the concept of the external power system
dynamic equivalent for the study system, this paper proposes a
novel evolutionary method for the identification of the
equivalent’s parameters, comparing the answers of the
complete network and its equivalent following the small
disturbances which emerged in the study system. The proposed
method is demonstrated and compared with the original system
using the 10 machines 39 buses New England test system. The
comparison shows that the proposed approach can preserve all
dynamic properties of the original network.
Index Terms-- Dynamic equivalents, Dynamic modeling,
Genetic algorithm, Nonlinear identification, Multi-machine
system, Parameter identification, Power system, Structure
preservation.
I. INTRODUCTION
E LECTRIC power systems have long been perceived and
exploited like national entities or regional areas. The
interconnections between such zones being mainly used with
the aim of help in case of failure in a nearby zone. In the
new context of the electricity sector, this situation has
changed since the interconnections are increasingly used in
order to maximize the exchange of electricity through power
exchanges recently introduced. This leads the transmission
system operators (TSOs) to operate their systems
increasingly close to their limits. In an individual way, each
TSO is obliged to re-examine its way of doing the usual
studies on its own network. In particular, a more refined
modeling of the influence of neighboring systems on its own
should be considered.
Moreover, the process of extension of the synchronous
area which continues today, led to a significant increase in
the size of dynamical systems to study. Besides the difficulty
posed by the size, new structural phenomena occur from this
extension as the electromechanical oscillations of low
frequency observed between generators remote network.
Many dynamic equivalence techniques have been
developed over the years [1]-[9]. More research focused on
coherency equivalents and modal equivalents nowadays,
however, they need detailed data of the external system
which may be difficult to get in the power market
environment. Some estimation methods are based on
1
R. Gueddouche, USTHB, Electrical & Industrial Systems Laboratory-
LSEI, BP.32 El-Alia, Bab Ezzouar 16111 Algiers, Algeria (e-mail:
gueddouche.r@gmail.com)
2
M. Boudour, USTHB, Electrical & Industrial Systems Laboratory-
LSEI, BP.32 El-Alia, Bab Ezzouar 16111 Algiers, Algeria (e-mail:
mboudour@ieee.org).
& M. Boudour2
available information from the boundaries nodes and do not
require any knowledge detail of the external network. The
classical estimation methods are mainly based on the
linearization of the system around an operating point with
theoretical constraints validity [10] - [12]. Consequently,
they are limited by their validity in nonlinear practical
applications, offline and online applications.
This paper proposes a new evolutionary approach for
estimating parameters to determine a dynamic equivalent
model of an external power system from synchronized
measurements of disturbance occurred in internal system,
which obtained by PMUs (Phasor Measurement Units). This
approach was applied to 10 machines 39 buses New England
test system [13]. Comparing the properties of the equivalent
system with ones of the original system, the result shows
that the reduced system can represent dynamic behaviors of
the original system well, for any kind of disturbance.
II. DYNAMIC EQUIVALENTS
The equivalent is represented by a model with unknown
parameters. In operation, the power system is often
perturbed by small random disturbances. The estimation of
unknown parameters of the equivalent model is operated by
comparing the measurements obtained from the real
network, and the same measurements made on the
equivalent network by minimizing some objective function.
A. Dynamic equivalent model
The basic requirement of the dynamic equivalents is that
the response of the equivalent system can fit the original
system dynamically and approximately when faults happen
in the internal system. In our equivalent system, equivalent
generator represents dynamic effect of the external system
on the internal system located at the boundary bus (Fig.1).
When there are multiple boundary buses connecting the
external system with the internal system, multiple generators
are adopted to represent dynamic influence of the external
system on the internal system, with an equivalent generator
located at each boundary bus [14].
Parameters to be estimated are: 𝑋𝑑 , 𝑋𝑞 , 𝑋 ′ 𝑑 , 𝐻, 𝐷, 𝑇 ′ 𝑑𝑜 ,
voltage regulator gain 𝐾𝑎 and its time constant 𝑇𝑎 .
 • ∆𝜔 Angular velocity rotor variation (pu).
• Δ𝐸 ′ 𝑞 Internal voltage variation (pu).
• Δ𝐸𝑓𝑑 Internal excitation voltage variation (pu).
• 𝛥𝑇𝑚 Mechanical torque variation (pu).
• 𝛥𝑈𝑠 Order excitation systems variation (pu).
• Δ𝑃𝑒 Active power variation (pu).
• 𝛥𝑉𝑡 Change in terminal voltage generators (pu).
The equations describing the dynamic equivalent system
are expressed similar to those of the original system (5), but
Fig. 1.a. Original System Fig. 1.b. Equivalent System with a simplified structure and less parameters.
Δ𝑥 𝛼 = 𝐴 𝛼 ∆𝑥 𝛼 + 𝐵∆𝑢 (6)
The equivalent generator model can be expressed by the Δ𝑦 𝛼 = 𝐶 𝛼 ∆𝑥 𝛼
following equations: 𝐴 𝛼 , 𝐶 𝛼 , ∆𝑥 𝛼 𝑒𝑡 Δ𝑦 𝛼 , are all functions of 𝛼, the
1
𝜔 = 2𝐻 (𝑇𝑚 − 𝐸′𝑞 𝐼𝑞 + 𝑋𝑞 − 𝑋′𝑑 𝐼𝑑 𝐼𝑞 − 𝐷(∆𝜔 − 1)) (1) equivalent system parameter vector to be identified.
Therefore, an error function may be defined by :
𝛿 = 𝜔0 (∆𝜔 − 1) (2)
𝑒 𝛼 = ∆𝑦 − ∆𝑦 𝛼 (7)
1
𝐸 ′𝑞 = 𝑇′ (𝐸𝑓𝑑 − 𝑋𝑑 − 𝑋′𝑑 𝐼𝑑 − 𝐸′𝑞 ) (3) Where ∆𝑦 represents the responses of the original system,
𝑑0
1 which are directly measureable.
𝐸𝑓𝑑 = 𝑇 (𝐾𝑎 𝑉𝑟𝑒𝑓 − 𝑉𝑡 + 𝑉𝑠 − 𝐸𝑓𝑑 ) (4) And ∆𝑦 𝛼 , must be calculated by simulating the
𝑎
equivalent system with the same disturbance.
B. Parameter identification of equivalent model 2) Objective function
The parameters identification for nonlinear systems The idea is to minimize the error function (7) between the
includes deterministic and evolutionary methods. measure and the model output for all machines belonging to
Evolutionary methods offer greater ease of adjustment the internal network. The mathematical model of the
problems and other advantages over deterministic methods optimization problem is a multi-objective function and can
because they: be formulated as follows:
- Require only the calculation of the objective function,
without this last being forced to be continuous or 𝑚𝑖𝑛 𝑒 𝛼 = 𝑒1 𝛼 , 𝑒2 𝛼 , … , 𝑒3 𝑚 −𝑛𝑒𝑥 𝛼
differentiable. (8)
- Easily adaptable to multi-objective problems 𝛼>0
optimization, and complex systems with very Where:
important unknown number of parameter. 𝑡 𝑓𝑖𝑛
- Have great theoretical probability to find the global 𝑒𝑖 𝛼 = ∆𝑦𝑖 𝑡 − ∆𝑦𝑖 𝛼, 𝑡 ; 𝑖 = 1,2, … ,3 𝑚 − 𝑛𝑒𝑥
optimum. 𝑡=𝑡𝑜
- Offer great simplicity of implementation. By weighting the measurements by a weighting factor 𝜆𝑖 ,
An evolutionary approach based on genetic algorithms is our objective function becomes:
presented in this document, to identify the dynamic
equivalent model parameters, by minimizing the range of 3 𝑚 −𝑛𝑒𝑥 𝑡 𝑓𝑖𝑛
𝑚𝑖𝑛 𝑒 𝛼 = 𝑖=1 𝜆𝑖 𝑡=𝑡𝑜 ∆𝑦𝑖 𝑡 − ∆𝑦𝑖 𝛼, 𝑡
speed variation, electrical power variation and terminal
voltage variation produced in all generators of the internal 𝛼>0
system between the original and the equivalent system. (9)
1) Error function Where:
The original system equations are generally written : 𝛼 = [𝑋𝑑 , 𝑋𝑞 , 𝑋 ′ 𝑑 , 𝐻, 𝐷, 𝑇 ′ 𝑑𝑜 , 𝐾𝑎 , 𝑇𝑎 ]
Δ𝑥 = 𝐴∆𝑥 + 𝐵∆𝑢 (5) ∆𝑦 =
𝑇
Δ𝑦 = 𝐶∆𝑥 ∆𝜔1 … ∆𝜔𝑚 −𝑛𝑒𝑥 Δ𝑃𝑒1 … Δ𝑃𝑒𝑚 −𝑛𝑒𝑥 Δ𝑉𝑡1 … Δ𝑉𝑡(𝑚 −𝑛𝑒𝑥 )
Where: ∆𝑦 =
A= State matrix (4m× 4m). 𝑇
∆𝜔1 … ∆𝜔𝑚 −𝑛𝑒𝑥 Δ𝑃𝑒1 … Δ𝑃𝑒𝑚 −𝑛𝑒𝑥 Δ𝑉𝑡1 … Δ𝑉𝑡(𝑚 −𝑛𝑒𝑥 )
B = Control matrix (4m×2m).
𝑚 = number of machines belonging to the complete system.
C = Output matrix (2m× 4m).
𝑚 − 𝑛𝑒𝑥 = number of machines belonging to the study
∆𝑥
𝑇 system (internal system).
= ∆𝜔1 … ∆𝜔𝑚 ∆𝛿1 … ∆𝛿𝑚 Δ𝐸 ′ 𝑞1 … Δ𝐸 ′ 𝑞𝑚 Δ𝐸𝑓𝑑1 … Δ𝐸𝑓𝑑𝑚 3) Encoding and initial population
∆𝑢 = 𝛥𝑇𝑚 1 … 𝛥𝑇𝑚𝑚 Δ𝑈𝑠1 … Δ𝑈𝑠𝑚 𝑇 The encoding of individuals is an important parameter in
∆𝑦 = ∆𝜔1 … ∆𝜔𝑚 Δ𝑃𝑒1 … Δ𝑃𝑒𝑚 Δ𝑉𝑡1 … Δ𝑉𝑡𝑚 𝑇 population research methods. These are represented as a
With : strings (chromosomes) containing characters or genes of a
• ∆𝛿 Absolute angle rotor variation (rad. elec/sec). predetermined alphabet. There are different ways to code a
solution. In our study, the individual is represented by eight Where :
parts of chromosome corresponding to the eight parameters 1
𝜆𝑖 = 𝑡 𝑓𝑖𝑛 ; 𝑖 = 1,2, … , 𝑁𝑚𝑒𝑠 ∗ 𝑚 − 𝑛𝑒𝑥
to be estimated, each gene (parameter) is represented by its 𝑡=𝑡𝑜 ∆𝑦 𝑖 𝑡
physical value, which means that, the real coding is adopted 5) Evolution mechanism
(fig.2).
The GAs requires an initial population to start search a) Parents selection
process. Applied methods generate randomly a set of Among initial population individuals, we choose :
solutions belonging to the following area: - A number of individual « 𝑁𝑚𝑒𝑠 ∗ 𝑚 − 𝑛𝑒𝑥 » ;
𝛼𝑚𝑖𝑛 ≤ 𝛼 ≤ 𝛼𝑚𝑎𝑥 (10) To form a group of (Local parents), in which each parent
Where: "i" satisfied the fitness function minimum:
𝑚 −𝑛𝑒𝑥 𝑚 −𝑛𝑒𝑥
𝑖=1 𝑋𝑑𝑖 2 𝑖=1 𝑋𝑑𝑖 𝑝𝑎𝑟𝑒𝑛𝑡𝑖 = min𝑗 =1..𝑁𝑖𝑛𝑑 𝑓𝑖 (𝐼𝑛𝑑𝑗 ) (13)
2 𝑚 −𝑛𝑒𝑥 𝑚 −𝑛𝑒𝑥
𝑚 −𝑛𝑒𝑥
𝑋𝑞𝑖 2 𝑚 −𝑛𝑒𝑥
𝑋𝑞𝑖
Where:
𝑖=1 𝑖=1
2 𝑚 −𝑛𝑒𝑥 𝑚−𝑛𝑒𝑥 𝑖 = 1,2, … , 𝑁𝑚𝑒𝑠 ∗ 𝑚 − 𝑛𝑒𝑥
𝑚 −𝑛𝑒𝑥
𝑖=1 𝑋 ′ 𝑑𝑖 2 𝑚 −𝑛𝑒𝑥
𝑖=1 𝑋 ′ 𝑑𝑖 - A number of individual « 𝑁𝑚𝑒𝑠 » ;
2 𝑚 −𝑛𝑒𝑥 𝑚 −𝑛𝑒𝑥 To form a group of (Global parents), in which each parent
𝑚 −𝑛𝑒𝑥 𝑚 −𝑛𝑒𝑥
𝑖=1 𝐻𝑖 2 𝑖=1 𝐻𝑖 "k" satisfied the fitness function minimum:
2 𝑚 −𝑛𝑒𝑥 𝑚−𝑛𝑒𝑥 𝑁𝑚𝑒𝑠 ∗ 𝑚 −𝑛𝑒𝑥
𝛼𝑚𝑖𝑛 𝑇 = 𝑚 −𝑛𝑒𝑥 ; 𝛼𝑚𝑎𝑥 𝑇 = 𝑚 −𝑛𝑒𝑥 𝑝𝑎𝑟𝑒𝑛𝑡𝑘 = min𝑗 =1..𝑁𝑖𝑛𝑑 𝑓𝑖 (𝐼𝑛𝑑𝑗 ) (14)
𝑖=1 𝐷𝑖 2 𝑖=1 𝐷𝑖 𝑖=1
2 𝑚 −𝑛𝑒𝑥 𝑚 −𝑛𝑒𝑥
𝑚 −𝑛𝑒𝑥 ′ 𝑚 −𝑛𝑒𝑥
𝑖=1 𝑇 𝑑𝑜𝑖 2 𝑖=1 𝑇 ′ 𝑑𝑜𝑖 Where: 𝑘 = 1,2, … , 𝑁𝑚𝑒𝑠
2 𝑚 −𝑛𝑒𝑥 𝑚 −𝑛𝑒𝑥
𝑚 −𝑛𝑒𝑥 𝑚 −𝑛𝑒𝑥
𝐾𝑎𝑖 2 𝐾𝑎𝑖
𝑖=1 𝑖=1
The local parents group includes the favorable solutions
2 𝑚 −𝑛𝑒𝑥 𝑚 −𝑛𝑒𝑥
𝑚 −𝑛𝑒𝑥
𝑇𝑎𝑖 2 𝑚 −𝑛𝑒𝑥
𝑇𝑎𝑖
for each type of measure that corresponds to each machine
𝑖=1 𝑖=1
2 𝑚 −𝑛𝑒𝑥 𝑚−𝑛𝑒𝑥 belonging to the internal system. However the global parents
group includes the favorable solutions for each type of
The individuals’ number in the initial population is chosen measure that corresponds to all the machines belonging to
such that: the internal system.
𝑁𝑖𝑛𝑑 > 𝑁𝑚𝑒𝑠 ∗ (𝑚 − 𝑛𝑒𝑥) (11) The population called, "Parents population" is made up of
Where: the two groups (Global parents & Local parents).
𝑁𝑚𝑒𝑠 = Number of measures nature considered
(𝜔, 𝑃𝑒, 𝑉𝑡 ) = 3. b) Multi-parent recombination
𝑚 = Total number of machines belonging to the complete Recombination is used mainly in evolutionary strategies.
network. Contrary to k points crossover operators, which exchange
𝑛𝑒𝑥 = Number of machines belonging to the external information between two parents, the recombination creates
system. the descendants, by weighting many parents components’.
A representation of Nind individuals (solutions) in an initial We define three weighting operators:
𝑃𝑟1 𝑐𝑕𝑟 +𝑃𝑟2 𝑐𝑕𝑟
population is shown in fig.2. 𝑂𝑝1 𝑐𝑕𝑟 =
2
𝑃𝑟1 𝑐𝑕𝑟 −𝑃𝑟2 𝑐𝑕𝑟
𝑂𝑝2 𝑐𝑕𝑟 = 𝑚𝑎𝑥 𝑃𝑟1 𝑐𝑕𝑟 , 𝑃𝑟2 𝑐𝑕𝑟 +
2
𝑃𝑟1 𝑐𝑕𝑟 −𝑃𝑟2 𝑐𝑕𝑟
𝑂𝑝3 𝑐𝑕𝑟 = 𝑚𝑖𝑛 𝑃𝑟1 𝑐𝑕𝑟 , 𝑃𝑟2 𝑐𝑕𝑟 −
2
Where:
𝑐𝑕𝑟 ∶ Integer belonging to [1,8], which represents crossed
chromosome order.
𝑃𝑟1 , 𝑃𝑟2 : Two individuals of local parents group.
We generate 𝑁𝑖𝑛𝑑 individuals for construct the new
population (generation) from the two groups' "local parents"
and "global parents", with the following steps:
Fig. 2. Organization and coding of individuals in initial population.
1. We reproduce the same individuals in the global parents
group for construct, 𝑁𝑖𝑛𝑑 individuals of the new
4) Evaluation mechanism
With a fitness function or adaptive function the evaluation population (Fig. 3).
is done in a closed interval 𝑡𝑜 , 𝑡𝑓𝑖𝑛 using each time a new
dynamic simulation of the equivalent system for each
population individuals.
𝑡 𝑓𝑖𝑛
𝑓𝑖 𝛼 = 𝜆𝑖 𝑡=𝑡𝑜 ∆𝑦𝑖 𝑡 − ∆𝑦𝑖 𝛼, 𝑡 (12)

Fig. 3. New population from the group Global Parents
2. For each individual "i" belonging to the new population:
- A random integer number of weighted point "k" is
generated between 1 and 8 ∗ 𝑇𝑑𝑖𝑣 .
- In each weightings j=1,2…,k :
- we choose a random pair of individuals from the
local parents group.
- a crossed chromosome is chosen randomly « chr ».
- a random real number, “h”, is generated between 0
and 1, to select the weighting operator (Fig. 4).
Fig. 4.Recombination operator distribution's
- the value obtained by the weighting operator is
assigned to the individual's chromosome "i(chr)" of
the new population, if this value satisfies the
constraint:
𝛼>0 (15)
- we move to the next individual "i +1" when the number
of weighting point "k" of the individual "i" is complete.
3. Accordance with the procedure for selection of parents,
the best individuals evaluated by the fitness function (12)
between parents population and this new population,
construct the new parents population and clear the old
parents population for the next generation (elitist
strategy).
III. ALGORITHM
The complete algorithm of the proposed method is given
below:
Step1 : Introduction of static and dynamic data of the
complete system.
Step2 : Power flow calculation, linearization, disturbance
choice and network dynamic simulation.
Step3 : Border node choice and equivalent system linear
model construction from results of the power
flow.
Step4 : Generating the initial population according to the
procedure specified in subsection (III.B.3).
Step5 : Run equivalent system dynamic simulation and
evaluate all individuals by fitness function (12).
Step6 : Construct the population of parents following the
procedure described in paragraph (III.B.5.a),
with equations (13,14).
Step7 : While the number of generation has not reached
the maximum number 𝑁𝑔𝑒𝑛 :
Step7.1 : Generate the new population using described
procedure in paragraph (III.B.5.b).
Step7.2 : Run equivalent system dynamic simulation
and evaluate all individuals by fitness
function (12).
Step7.3 : Select the new individuals in parents
population following described procedure
in paragraph (III.B.5.b.3).
Step8 : End while.
Step9 : Select best individuals in parents population with
the objective function (9), and display the
results.
Table.I, represents the most significant parameters that
characterize this algorithm.
TABLE I
CONTROL PARAMETERS
Control parameters Definition
𝑁𝑖𝑛𝑑 Number of individual in a population
𝑁𝑔𝑒𝑛 Maximum number of generation
𝑇𝑑𝑖𝑣 Weighting rate
𝑟𝑒𝑐1 , 𝑟𝑒𝑐2 , 𝑟𝑒𝑐3 Recombination rate
𝑂𝑝1 , 𝑂𝑝2 , 𝑂𝑝3 Recombination operator
IV. APPLICATION
To validate the proposed approach, developed algorithm
was used to build the dynamic equivalent of transmission
network IEEE New-England 39 bus [13]. It represents a
simplified New England transmission network (northeastern
United States) that is part of real U.S. network. This network
consists of 10 generators (PGtotal= 6.19 GW, QGtotal= 1.28
GVAr) and 39 bus with 19 load bus. The original system
structure is as Fig.5.
Fig.5.Original system IEEE-39 bus
 In Fig. 5, the shaded part is the internal system and the about 12 cycles. Time domain dynamic simulations were
rest is the external system to be reduced. There are a performed both on the two systems.
boundary bus 16 and two tie lines 16-15 and 16-17 between
the internal and the external systems. Fig. 6 is the sketch of TABLE III
EQUIVALENT GENERATOR PARAMETERS
the equivalent system.
Parameter
Dist. 𝑿𝒅 𝑿𝒒 𝑿′ 𝒅 𝑯 𝑫 𝑻′ 𝒅𝒐 𝑲𝒂 𝑻𝒂
0.471 0.043 0.060 135.0 2.5 7.898 155.0 0.068
3
8 8 3 9 ∙ 10−5 6 8 5

Dynamic responses are shown in fig. 7 to fig. 9. The

boundary node voltage is shown by fig. 7, The angle of the
machine n°10 is shown in fig. 8 and angular velocity of the
machine n°8 in fig. 9.

Fig. 6. Equivalents system

A. Simulation
The network linear model is used by the algorithm for
solutions evaluation in estimation procedure. The proposed
algorithm programming and dynamic linear model
simulation was implemented in Matlab.V7 environment.
The algorithm control parameters are summarized in
Table II. These parameters were obtained after several tests
with an appropriate adjustment. Fig. 7. Voltage variation at bus 16
TABLE II
CONTROL PARAMETER VALUES
Control parameters Values
𝑁𝑖𝑛𝑑 80
𝑁𝑔𝑒𝑛 60
𝑇𝑑𝑖𝑣 (%) 100
𝑟𝑒𝑐1 , 𝑟𝑒𝑐2 , 𝑟𝑒𝑐3 0.5 , 0.25 , 0.25

To allow a better parameters estimation, different

disturbances were applied in different parts of the internal
system. Each proposed disturbance has a period of 10s. An
event is generated each time to the first second of the
simulation. The algorithm verifies the equivalents behavior
in period of 10s, by comparing its response with the
response of the original system. If the equivalent response
exceeds some level relatively to the original systems
response in a new disturbance, the equivalent generator
parameters are recalculated and updated. Each solution
obtained by applying the previous disturbances must be part
of the initial population of the next estimation procedure.
The estimated parameters of the equivalent generator are
presented in Table III.
B. Validation
A small perturbation represented by disconnecting the line
3-18 is applied to bus18 of the equivalent system and the
original system respectively at time, t=2s and lasted for
Fig. 8. Angle of the machine 10
In order to verify the performance of the proposed
dynamic equivalent approach in the case of the large
disturbances, three-phase fault was applied to bus 30 of the
equivalent system and the original system respectively at
time t=2s, and the fault lasted for about 3 cycles. Time
domain dynamic simulations were carried out on the original
system and the equivalent system, and dynamic responses
are shown in fig. 10 and fig. 11. The angle of the machine
n°10 is shown in fig. 10 and angular velocity of the machine
n°8 in fig. 11.

Fig. 9. Angular velocity rotor variation of machine 8
Fig. 10. Angle of the machine 10
Fig. 11. Angular velocity rotor variation of machine 8
The simulation results of equivalent system with
estimated parameters are encouraging and show the
effectiveness of the developed algorithm and the ability of
the estimated equivalent to reproduce the influence of the
external system on internal system for small and severe
disturbances.
V. CONCLUSION
In this study, we proposed a nonlinear estimation method
for parameters, based on an evolutionary algorithm. In order
to obtain the external system dynamic equivalent, several
dynamic simulation disturbances were applied to increase
the accuracy of the model estimated by the developed
algorithm.
The proposed approach does not require data of external
system. This approach requires only configuration
information, settings and operating status of the internal
system.
Dynamic simulation was performed on both the original
system and the equivalent system under different operating
conditions, and the results show that the obtained equivalent
system can represent the main dynamic characteristics of the
original system well. Thus, the approach proposed is proved
to be feasible and has potential for tackling the complex
practical application in power system.
VI. RÉFÉRENCES
[1] Wang Lei, M. Klein, S. Yirga, and P. Kundurl, "Dynamic reduction of
large power systems for stability studies," IEEE Trans. Power
Systems, vol. 12, no. 2, pp. 889-895, 1997.
[2] M. L. Ourari, L. A. Dessaint, and D. Van-Que, "Dynamic equivalent
modeling of large power systems using structure preservation
technique, " IEEE Trans on Power Systems, vol. 21, no. 3, pp. 1284-
1295, 2006.
[3] Sung-Kwan Joo, Liu Chen-Ching, L. E. Jones and Jong-Woong Choe,
"Coherency and aggregation techniques incorporating rotor and
voltage dynamics," IEEE Trans. Power Systems, vol. 19, no. 2, pp.
1068-1075, 2004.
[4] W. W. Price, E. M. Gulachensk, "Testing of the modal dynamic
equivalents technique," IEEE Trans. Power Apparatus and Systems, ,
vol. PAS-97, no. 4, pp. 1366-1372, 1978.
[5] K. Nojiri, S. Suzaki, K. Takenaka and M. Goto, "Modal reduced
dynamic equivalent model for analog type power system simulator,"
IEEE Trans. Power Systems, vol. 12, no. 4, pp. 1518-1523, 1997.
[6] Yu Yaonan, M. A. El-Sharkawi, "Estimation of external dynamic
equivalents of a thirteen system," IEEE Trans. Power Apparatus and
Systems, vol. PAS-100, no. 3, pp. 1324-1332, 1981.
[7] W. W. Price, D. N. Ewart, E. M. Gulachenski and R. F. Silva,
"Dynamic equivalents from on-line measurements," IEEE Trans on
Power Apparatus and Systems, vol. 94, no. 4, pp. 1349-1357, 1975.
[8] J. M. Ramfirez, R. G. Valle, "A technique to reduce power systems
electromechanical models," IEEE Transa on Energy Conversion, vol.
19, no. 2, pp. 456-458, 2004.
[9] Yang Jingping, Xu Zheng. "Application of Dynamic Equivalents
Based on Identificaiton of Coherent Generator Group in Engineering,"
Power System Techonology, vol. 29, no. 17, pp. 68-71, 2005.
[10] IBRAHIM M A H, MOSTAFA O M, El-ABIAD A H. "Dynamic
Equivalent Using Operating Data and Stochastic Modeling". IEEE
PAS, 1976, 95(5):1713-1722.
[11] ALVARADO F L. "Real Time External Equivalents for Static
Security Analysis". IEEE TPAS, 1979, 98: 505-507.
[12] YU Y. N. "Power System Dynamics. Academic Press", 1983.
[13] J.E. Fagan."Synchronous Machine Modeling Mechanization and
System Performance Study". Thèse de doctorat, The University of
Texas at Arlington, mai 1977.
[14] J. M. Ramirez Arredondo. " Obtaining dynamic equivalents through
the minimization of a line flows function". International Journal of
Electrical Power & Energy Systems, vol. 21: 365-373, (1999).